Forward pass while using auxilary loss

vision
1 
def forward(self, x):
        """
        Inputs:
        - x: PyTorch input Variable
        """
        return self.backbone(x)['out']

def forward(self, x):
        """
        Inputs:
        - x: PyTorch input Variable
        """
        return self.backbone(x)['aux']

I want to use auxilary loss while doing semantic segmenation. But which one should I return ‘out’ or ‘aux’? In the documentation it is stated that

class DeepLabV3(SimpleSegmentationModel):
“”"
Implements DeepLabV3 model from
"Rethinking Atrous Convolution for Semantic Image Segmentation" <https://arxiv.org/abs/1706.05587>.

Arguments:
    backbone (nn.Module): the network used to compute the features for the model.
        The backbone should return an OrderedDict[Tensor], with the key being
        "out" for the last feature map used, and "aux" if an auxiliary classifier
        is used.
    classifier (nn.Module): module that takes the "out" element returned from
        the backbone and returns a dense prediction.
    aux_classifier (nn.Module, optional): auxiliary classifier used during training
"""
pass